Three jokes about the same news.
Every morning, our agent picks the day’s top AI story and asks three editorial cartoonists what they make of it. The first is bleak; the second is breezy; the third is a New York Times op-ed in a black tie. Same event, three incompatible drawings — the rest of the site in poster form.
No human cartoonist was harmed.
Each morning, after the daily agent run, we take the most recently processed article and the three editorial summaries written for it. DeepSeek then writes a one-paragraph “cartoon brief” for each voice — central figure, visual metaphor, props, expression. Each brief is appended to a locked style anchor (the same wording every time, hand-tuned per voice) and rendered by OpenAI’sat 1024×1536. The resulting PNGs land in Supabase Storage and the metadata in our cartoons table — nothing is ever deleted.
The style anchors are deliberately locked: changing one word changes the look of the whole archive. The Dark anchor pulls toward Steve Bell and Martin Rowson; the Light anchor toward Mary Blair’s mid-century gouache; the Neutral anchor toward New York Times op-ed line art.
These aren’t hand-drawn cartoons. They are an AI’s read of the AI’s read of the news. That is, slightly disconcertingly, the point.